Ballerburg9005 has quit [Ping timeout: 252 seconds]
Ballerburg9005 has joined #linux-amlogic
Ballerburg9005 has quit [Ping timeout: 255 seconds]
nashpa has quit [Ping timeout: 265 seconds]
hexdump0815 has quit [Ping timeout: 260 seconds]
hexdump0815 has joined #linux-amlogic
dliviu has joined #linux-amlogic
buzzmarshall has quit [Quit: Konversation terminated!]
dliviu has quit [Ping timeout: 248 seconds]
dliviu has joined #linux-amlogic
naoki has quit [Ping timeout: 252 seconds]
naoki has joined #linux-amlogic
dliviu has quit [Ping timeout: 248 seconds]
dliviu has joined #linux-amlogic
vagrantc has quit [Quit: leaving]
naoki has quit [Ping timeout: 255 seconds]
naoki has joined #linux-amlogic
camus has quit [Read error: Connection reset by peer]
camus has joined #linux-amlogic
camus has quit [Quit: camus]
Ballerburg9005 has joined #linux-amlogic
chewitt has joined #linux-amlogic
<chewitt>
phh: I'm interested to know if the HEVC content that you're playing with your changes is 8-bit or 10-bit?
<phh>
chewitt: hevc is not working
<phh>
possibly because i tried only 10bit content
<chewitt>
on g12a there's an unresolved bug where playing 8-bit HEVC works, but 10-bit wedges the board
<chewitt>
(which I would love to find a solution to)
<chewitt>
there's also a bug with VP9 and content with 29.97/59.94 fractional rates: "[ 469.650609] Fatal Error, invalid HDMI vclk freq 593406"
<chewitt>
something to be aware of if nothing else .. it sucks to test and hit bugs that aren't your own creation :)
<phh>
right thanks. atm i'm not really trying to make it fully working/stable. for the moment i just want to explore what's supported mainline on that platform compared to vendor kernel
<phh>
chewitt: k, playback of hevc 8 bit looks okay (tested with `ffmpeg -c:v hevc_v4l2m2m -y -i jellyfish-3-mbps-hd-hevc.mkv -c:v rawvideo -f nut - | mpv --vo=tct --ao=null -` with debian testing packaged ffmpeg), but meson vdec shouts "Couldn't find ref. frame 0"
<chewitt>
it has quite a few improvements for v4l2_m2m
<chewitt>
this is the ffmpeg source for RPi boards which have v4l2_m2m H264 IP (while HEVC is v4l2_request)
<phh>
it's a completely hw component doing h264 decoding from h265 on rpi4?
<chewitt>
i've been working with them for a while to ensure the stuff that is being prepped for upstreaming to ffmpeg still works on other SoC platforms
<chewitt>
RPi4 is basically the same decoders as RPi3 but with an extra IP block from Broadcom to HEVC bolted on the side
<chewitt>
RPi4 is a little unusual in that it mixes both stateful/stateless methods
<phh>
splendid
<chewitt>
unfortunately the 10-bit issue in HEVC has nothing to do with ffmpeg (else it would have been fixed)
<phh>
so yeah, my hevc 8bit decoding is broken just like my vp9 8bit, you can see an example at https://phh.freeboxos.fr:49026/share/7EUrdhoaqHy7loIW/hello.nut ; basically I get two correct frames, then pixels get uh, mangled? like global color is still correct
<phh>
that's with debian's ffmpeg though. I'll try the branch you mentioned soon (but i'm building on target, so it takes a lot of time)
<phh>
h264 uhd decoding i'd say. but i can't guarantee it was actually fixing something and not that i was hitting another issue
<phh>
(like if the decoder is stuck, maybe the esparser will fill vififo and trigger vififo error just because the decoder doesn't consume it)
<phh>
(or it was when i added a shitload of logs and hw decoding was output 0.3fps because of all the logs)
f_ has joined #linux-amlogic
f_ has quit [Ping timeout: 246 seconds]
f_ has joined #linux-amlogic
buzzmarshall has joined #linux-amlogic
vagrantc has joined #linux-amlogic
naoki has quit [Quit: naoki]
Daaanct12 has quit [Quit: Quitting]
Danct12 has joined #linux-amlogic
Ballerburg9005 has quit [Ping timeout: 248 seconds]
Daanct12 has joined #linux-amlogic
Danct12 has quit [Ping timeout: 252 seconds]
Ballerburg9005 has joined #linux-amlogic
Daaanct12 has joined #linux-amlogic
Daanct12 has quit [Ping timeout: 252 seconds]
f_ has quit [Ping timeout: 268 seconds]
<narmstrong>
H264 is broken on g12a, a new h264 decoder was started by using the same HW as vp9 & hevc but was never finished
<narmstrong>
The good point we we could bypass ESPARSER and directly feed buffers to the FW (it’s also possible on vp9 & hevc) but there’s nobody available to finish all this…